Georges, Sorry for the lengthy delay. Beamline, travel, etc -- I just haven't had a chance to look into software issues until this week. I agree that the fits are being evaluated slightly differently in the old and new versions of the code. As you say, best fit values and statistical parameters are somewhat different. I am not sure I agree that they are "nowhere near similar", but the disrepancy is much larger than a mere numerical difference. I am not yet clear what is going on, but I am looking into it. If I may comment a bit more broadly on the project you sent me, I have some concerns about your fitting model. I would not agree that your fit is "good" using either version of Artemis. Or perhaps I should say that you and I may not have the same expectation of what "good fit" means. The central problem in your fitting model is that you are using the same parameter for sigma^2 for all your paths, which include O, C, and N atoms spread over a distance range from 1.69 A to 2.03 A. Similarly, you use the same deltaR parameter for all four paths. I understand that you have rather limited information (although you have usable data out to at least 13 invAng), but I don't really accept your constraint that the O atom at 1.69 should have the same sigma^2 as the N atom at 2.04. My intuition tells me that the short O atom should be very tightly bound to the tungsten absorber. Consequently, the O should have a much smaller sigma^2 than the N. So, I question whether your model is defensible, thus I dispute that the one fit is "good" while the other is "bad". That said, they are significantly different. You have uncovered a problem that needs to be figured out and fixed. I'll keep you posted. B On Monday, April 22, 2013 04:05:46 PM Georges Siddiqi wrote:
Hi All,
I've done a fit in the old version of Artemis (0.8.014), and wanted to import this fit into the new version of artemis (demeter 0.9.15) that I have to play around with it a bit more.
The problem is, when I import the the "old-style artemis project", and do the same fit, the values, and fit itself are nowhere near similar to what i got with the old version of Artemis.
what is the cause of this discrepancy? As far as I can tell from double checking all the fitting parameters, windows, variables, etc, they are all exactly the same. Even if I try to just do a summation in the new version with my best-fit values from the old version, the fit is completely different.
Previously when I was importing fits between versions I had no issues whatsoever.
I've attached the artemis project file (W EXAFS.apj) that gives me a good fit in the old version, and a poor fit in the new version.
thanks, georges
-- Bruce Ravel ------------------------------------ bravel@bnl.gov National Institute of Standards and Technology Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973 Homepage: http://xafs.org/BruceRavel Software: https://github.com/bruceravel